Nicolás Cánovas, General Manager, AMD for Latin America, describes how the concepts of Machine Learning and Deep Learning have evolved.
Currently, Artificial Intelligence (AI) has become a highly popular phenomenon, positioning itself as a key technology to drive progress in various fields such as healthcare, automotive, logistics, customer service, data analysis, cybersecurity and many others.
AI is mainly used as a generic term for all forms of computation-based intelligence. In general, it applies to any system that imitates human processes of learning and decision-making in response to information, data analysis, pattern recognition or strategy development.
The terms Machine Learning (ML) and Deep Learning (DL) better describe the reality of current intelligent computer systems and the problems they can solve for developers and end-users.
Public awareness of AI increased significantly with the launch of ChatGPT, Microsoft Bing Chat and Google Bard in late 2022. Since then, the volume of news has skyrocketed by almost 500%, for example, going from 4,700 articles in October in US media to nearly 22,000 in April.
However, AI does much more than enhancing generative language models. In fact, at AMD, we have driven its use to personalize the performance of Ryzen Processors to offer the necessary efficiency and adapt to the modern needs of our users.
Focus on Innovation
Our role in this virtuous circle of innovation translates into providing high-performance computing technology to support the vast amount of information required by any AI application, no matter how simple.
We work to promote and expand the capabilities and optimizations of ML. After acquiring Xilinx, AMD has propelled the Vitis AI tool, thus providing a comprehensive platform for AI inference development on Alveo data center accelerators and AMD adaptable SoCs.
Vitis AI connects to common software development tools and utilizes a wide range of open-source libraries optimized to empower developers with machine learning acceleration.
Similarly, AMD Ryzen Series 7040 Processors with Ryzen AI feature a chip specifically designed to perform all kinds of activities according to each customer’s requirements. The performance of AMD hardware and associated software also provides significant benefits to the development and testing process of Artificial Intelligence systems.
Currently, a computing platform based on AMD’s latest technologies (AMD EPYC CPU and Radeon Instinct GPU) can develop and test a new intelligent application in days or weeks, a process that used to take years.
Understanding Machine Learning and Deep Learning
The current state-of-the-art ML and DL computer intelligence systems can adjust operations after continuous exposure to data and other types of information. While they are related in nature, there are subtle differences that set these fields apart within computer science.
ML refers to a system that can actively learn by itself, rather than just passively receiving and processing information. The computer system is coded to respond to the given information as if it were human, using algorithms that analyze the data for patterns or structures. ML algorithms are designed to improve performance over time as they are exposed to more data.
When a human recognizes something, that recognition happens instantly. To help mimic this process, ML algorithms use neural networks. Just like the human learning process, the computation in neural networks classifies data based on recognized elements within the image.
The success rate of correct classification can improve over time through feedback provided by ‘expert’ humans, helping the system learn and discern correct decisions from incorrect ones, aiming for maximum efficiency and greater accuracy. The neural network algorithm adjusts all future decisions based on the received feedback. This process mimics human recognition by training the network to produce the desired outcome.
A neural network could make a statement, and then the algorithm applies this learning to the data, seeking and categorizing the defined elements. This process can improve over time with the help of information provided by individuals.
The neural network algorithm adjusts all future decisions based on the received information, resulting in more accurate data collection.
For example, if the provided information were that ‘every shape has various variations,’ the algorithm could organize the results as follows: Google hired professional photographers and documentary specialists to provide technical guidelines to train the neural network-based algorithm behind their smart camera, Clips.
The information provided helped the camera become more intuitive not only in the technical aspects of digital photography but also in anticipating more abstract qualities in capturing memorable moments.
DL focuses on a subset of ML that goes even further to solve problems, inspired by how the human brain recognizes and remembers information without external input from experts to direct the process.
DL applications must access vast amounts of data from which they learn. DL algorithms use deep neural networks to access extensive sets of information, explore and analyze them—for example, all music files on Spotify or Pandora to make continuous music suggestions based on a specific user’s preferences.
The main distinguishing factor between DL and ML is the representation of data. For instance, in the aforementioned example of Google’s Clips ML camera, input from professional photographers was needed to train the system.
However, in DL systems, experts are not required for precise feature identification. The data, whether an image, a news article, or a song, is evaluated in its natural, unprocessed form with minimal transformation. This process of unsupervised training is sometimes referred to as representation learning. During training, the DL algorithm progressively learns from the data to improve the accuracy of its conclusions (also known as inferences).
